evhub comments on Gears-Level Mental Models of Transformer Interpretability